Did English Generators Play Cournot? Capacity withholding
in the Electricity Pool.
by
04-010 March 2004
Richard Green
WP
Did English Generators Play Cournot?
Capacity withholding in the Electricity Pool.
Richard Green*
University of Hull Business School
University of Hull
Hull HU6 7RX
Tel: 01482 465720
Fax: 01482 466216
Email: R.J.Green@hull.ac.uk
Presently visiting MIT Center for Energy and Environmental Policy Research
E40-407, MIT CEEPR
1 Amherst Street
Cambridge MA, 02139-2347
March 2004
Revised: 6 April, 2004
Abstract
Electricity generators can raise the price of power by withholding their plant
from the market. We discuss two ways in which this could have affected
prices in the England and Wales Pool. Withholding low-cost capacity which
should be generating will raise energy prices but make the pattern of
generation less efficient. This pattern improved significantly after
privatisation. Withholding capacity that was not expected to generate would
raise the Capacity Payments based on spare capacity. On a multi-year basis,
these did not usually exceed “competitive” levels, the cost of keeping stations
open. The evidence for large-scale capacity withholding is weak.
Keywords: Electricity prices, Cournot competition, capacity withholding
JEL: L94
*
Support from the Economic and Social Research Council, under project R000 23 6828, Developing
Competition in the British Energy Markets, and from the Leverhulme Trust, through the award of a Philip
Leverhulme Prize, is gratefully acknowledged. I would like to thank the MIT Center for Energy and
Environmental Policy for its hospitality, Ijaz Rasool for help with the data, and Joanne Evans and Paul Joskow
for helpful comments.
1. Introduction
Few people would dispute that when an electricity industry is liberalised, market power can
be a problem. Incumbent generators frequently hold a high proportion of capacity, imports
are restricted by limits on the transmission lines to adjacent markets, demand is inelastic, and
the product cannot be stored. The term “withholding” is often used to describe the way in
which generators could exploit market power. “Economic withholding” implies that a plant
would not offer its output as soon as the market price was high enough to cover its costs of
doing so, but would wait until the price had risen above its costs. “Physical withholding”
implies that the plant’s output is not made available to the market at any price. In both cases,
the plants that are withheld from the market will generate less, and are likely to make less
money, but the strategy can increase the company’s profits by raising the price received by its
other units.
The strategy of withholding units from the market is a close fit to the economists’
model of Cournot competition, which is the most popular model of market power in
electricity. In a Cournot model, firms decide how much output to sell in each time period,
and prices are set by the intersection of the demand curve with this quantity. Cournot models
can make use of the rich cost data that is usually available, and generally produce a unique
equilibrium. These attractive features might be offset, however, by the fact that most
electricity markets are built around auctions in which generators offer prices for tranches of
output, and the fixed output assumption of the Cournot model seems empirically
inappropriate.
It might be possible to rescue the Cournot approach by pointing out that generators
need not offer all of their capacity to the market in every period. Kreps and Scheinkman
(1983) have shown that when generators choose capacities and then compete in prices, the
process can end up with prices at the Cournot level. If generators can be sure that they will
collectively offer no more capacity than the market will need, then they can drive up prices
without fearing that their rivals will have the spare capacity to expand output and steal their
market share.
This paper asks whether British generators have been keeping capacity away from the
market in order to raise prices above competitive levels. It examines two possible strategies.
One is to reduce the overall level of capacity available, keeping high-cost plants from the
market, which will raise the component of the British spot price designed to reward capacity
when it is scarce. This strategy was described by the regulator (Offer, 1991) and studied by
Newbery (1995). The second strategy is to reduce the overall level of capacity by keeping
low-cost plants from the market, ensuring that a higher-cost station sets the energy
component of the spot price, at a higher level than if all the capacity was available. This
strategy is discussed by Wolak and Patrick (1997), who support their argument with data
implying that capacity was being withheld at stations owned by the dominant British
generators, which had lower availability than comparable US plants.
1
This paper argues that availability figures do not provide conclusive evidence of
capacity withholding to raise prices. Capacity could be held back from the market at times
when there was no likelihood of it being called on to generate, with no impact on prices. The
stations would not be commercially available (the definition used in the British electricity
market) even though, technically, they were not subject to maintenance and could be made
available (a more traditional definition, and that used in the US). To look for evidence of
capacity withholding related to market power, we must examine prices and generation
patterns.
The next section of the paper outlines the British electricity market. The third section
considers power markets and approaches to modelling them. Section 4 discusses “simple”
capacity withholding, aimed to drive up the capacity payment. Section 5 discusses
“sophisticated” withholding aimed at the energy price. A final section offers brief
conclusions.
2. The British Electricity Market
The electricity industry in England and Wales was restructured in March 1990. The
integrated Central Electricity Generating Board was divided into three generating companies
and the National Grid Company (NGC), responsible for transmission. NGC also operated the
daily spot market, the Pool, in which practically all electricity had to be traded. Every day,
generators submitted price bids for each of their available units, and NGC calculated the
least-cost operating schedule for the following day. The System Marginal Price (SMP) for
each half-hour was based on the bid of the most expensive unit in normal operation during
that half-hour. A second component of the spot price, the capacity payment, was based on
the balance between available capacity and expected demand. NGC calculated the loss of
load probability (with some smoothing) and multiplied it by the value of lost load (set at
£2/kWh in 1990, and increased annually by the rate of inflation), less SMP, to give the
capacity payment. SMP and the capacity payment gave the Pool Purchase Price, paid to all
generators for their scheduled output. Generators that were available, but were not
scheduled, also received a capacity payment, and the cost of this, and various other payments
related to system support, were recovered through a charge known as Uplift.1
The Pool
Purchase Price plus Uplift gave the Pool Selling Price, paid by all buyers.
The main buyers in the Pool were the twelve Regional Electricity Companies (RECs),
responsible for distribution, and with an 8-year franchise over supply (retailing) to the
industry’s smaller consumers. The largest consumers, however, could choose their supplier
from the time of the restructuring, and the majority soon did so. Some bought from other
RECs, and a few of the largest bought on their own behalf in the Pool, but the most active
suppliers were the two major generators, National Power and PowerGen. These companies
1
In 1997, some of the charges recovered through Uplift were transferred to a new charge, the Transmission
Services Use of System charge. This is not formally part of the Pool Selling Price, and should be added to it in
any comparison of selling prices over time, but the distinction is not relevant to this paper.
2
had inherited all the industry’s conventional capacity (29 GW and 20 GW respectively),
while the nuclear stations were given to Nuclear Electric. They had proved impossible to
privatise at the start of the restructuring, although the more modern stations were eventually
privatised in 1996.
In the attempt to privatise the nuclear stations, however, the government had only
created two large conventional generating companies, and it soon became clear that they had
substantial market power. The regulator2
issued a series of reports on their behaviour,
culminating with a series of Undertakings announced in 1994. The companies agreed to sell
6 GW of capacity (about 15% of their total at the time), and to keep Pool prices below
specified levels in 1994/5 and 1995/6. All the capacity was sold to the same company
(Eastern, one of the RECs), and market power soon resurfaced as an issue. National Power
and PowerGen agreed to sell more capacity in 1998, in return for permission for vertical
mergers,3
which had been denied them in 1996. The companies have since sold more of their
capacity, on a voluntary basis. Nevertheless, the regulator tried to impose a “good behaviour
condition” in the regulatory licences4
of the eight largest generators. Two of the generators
resisted, and after the Competition Commission determined that the clause might be expected
to act against the public interest, it was removed from the licences of those generators that
had accepted it. Furthermore, the Pool was abolished and replaced by New Electricity
Trading Arrangements in March 2001, which the regulator and the government believed
would be less susceptible to market power.5
Divestitures by the incumbents were not the only way in which they lost market share
over the 1990s. Most of the RECs, and many later entrants, invested in new Combined Cycle
Gas Turbine stations, which have now taken nearly 20% of the market. The earlier stations
were all supported by long-term contracts, which hedged the Pool price. National Power and
PowerGen also sold much of their output under contract, which reduces their short-term
incentive to raise the spot price. High or volatile spot prices will increase the price at which
they can sell contracts in future, however, and so being heavily contracted does not eliminate
the companies’ incentive to exercise market power in the spot market.
National Power and PowerGen also invested heavily in new CCGT plant. In total,
18.5 GW of new CCGT capacity and 1.2 GW of nuclear capacity were added between March
1990 and March 2000. Since the system had spare capacity in 1990, and peak demand has
2
The Director General of Electricity Supply was responsible for regulating the industry. His office was
originally known as Offer (Office of Electricity Regulation) but combined with the gas regulator’s office in
1999 to form Ofgem (Office of Gas and Electricity Markets), with the same person appointed to regulate both
gas and electricity. In 2001, he became the first Chairman of the Gas and Electricity Markets Authority, which
has replaced the individual regulators.
3
PowerGen bought East Midlands Electricity, while National Power bought the supply business, but not the
distribution network, belonging to Midlands. All of the ex-REC supply businesses are now part of vertically
integrated groups.
4
Much of the detail of the industry’s regulation is governed through licences. All companies in the industry are
required to hold an appropriate licence for each of their activities. These can normally only be changed with the
company’s agreement, or if the regulator gains the support of the Competition Commission.
5
Not all independent commentators agree with this view (Hogan, 2000; Newbery, 1998; Wolfram, 1999). The
National Audit Office, however, concluded that NETA had “facilitated” lower wholesale prices (NAO, 2003,
para 6).
3
grown only from 48.3 GW to 51 GW, the incumbents have obviously been able to close a
large amount of capacity without risking shortages. Nevertheless, the regulator has been
concerned that the companies should not close an excessive amount of capacity, and the
companies agreed in 1992 to allow an independent assessor to examine their closure
decisions, should the regulator request this. This agreement was embodied in a new licence
condition that also required the companies to give the regulator an annual statement of their
plans for making each unit available to the market, a half-yearly update, and an end-of-year
reconciliation, noting and explaining any significant changes. The licence condition does not
lay down what the availability policies should be, but the regulator has enough information to
decide whether the companies are following their stated policies, and assess their impact on
the market.
The companies’ price bids are placed in the public domain a few days after they are
made, and the regulator has often studied them in reports on competition. In 1992, the
regulator agreed that stations that were used to meet constraints were justified in bidding at a
(high) level that nevertheless did not recover more than their costs over the year as a whole, if
they were only occasionally called on to generate at that level (Offer, 1992). In the mid-
1990s, the companies agreed to bid in such a way as to produce prices below a given level (in
normal circumstances). Apart from this, the regulator did not attempt to place other restraints
on the companies’ bids until the introduction of the “good behaviour” licence condition. He
commented on the level of prices, but tended to place less emphasis on the means used to
achieve that level.
3. Power Markets and Models
Economists have been interested in the ways electricity companies can raise prices above
their costs. Most electricity markets are based on some form of auction. In real time, the
system operator must match demand and supply, and the price bids from an auction allow
this to be done in an economically efficient manner. In the longer term, bilateral contracts
are common, but most markets have some forum in which buyers and sellers can sell power
on a short-term basis. In the Nordic countries, this forum is NordPool, while national
regulation markets are responsible for minute-by-minute balancing. In California, the Power
Exchange was one of several Scheduling Co-ordinators that match expected generation and
demand in advance, while the Independent System Operator dealt with short-term
adjustments and reserve. In England and Wales, the Pool was responsible for both day-ahead
trading and real-time adjustment, but the New Electricity Trading Arrangements that replaced
it separate the two. In most cases, however, the day-ahead market (or equivalent) provides a
marker price for the industry, and one set “as if” in a first-price auction.
In a perfectly competitive market, all generators would bid their capacity into this
market whenever they thought that there was a reasonable chance that it might be needed,
and would do so at a price equal to their marginal cost. If no generator owns more than one
unit, setting a higher price runs the risk that the unit would not be called, without affecting
4
the price that the unit would receive, which would be set by another generator almost all of
the time. A generator that does not offer its unit into the market cannot earn anything from it.
Only if there is very little chance that the unit would be required should it be withheld, in
order to save the costs of making it available needlessly. While most of the unit’s costs are
either fixed on a long-term basis (capital costs are sunk, while a basic level of staffing and
age-related maintenance are needed for as long as the station is open) or directly related to
output (fuel costs and maintenance based on the station’s operating regime), some staff costs
would be incurred if the station was made available but not required to generate.
A generator that owns many units and wishes to increase the market price has two
main alternatives. The first is to bid all of its available capacity into the market, but at prices
above its marginal costs. The second is to withhold some of its capacity from the market. If
the capacity should have been infra-marginal (i.e. it had low costs), this would mean that a
more expensive unit is now at the margin, raising the price. In England and Wales,
withdrawing extra-marginal capacity increases the capacity payment, but withdrawing extramarginal
capacity might have little impact in other markets, unless other generators interpret
the move as a signal to adopt higher-priced strategies.
The first papers on market power in England and Wales looked at bid-based price
increases. von der Fehr and Harbord, 1993 used an auction model to suggest that generators
would follow a mixed strategy with expected bids above marginal costs, although their
approach did not allow for detailed modelling of the generators’ capacities and cost
structures. They were able to show generators’ bid curves that were consistent with their
approach. Green and Newbery (1992) used the supply function approach, which can cope
with more detailed information on capacities and costs, but yields multiple equilibria. The
higher-priced equilibria involve supply functions that diverge from marginal cost at an
increasing rate as marginal cost itself rises. Green (1994) compares generators’ bids with
their marginal costs, and shows that they appear to be consistent with the supply function
approach, but does not use any formal statistics. Wolfram (1998) does use a statistical
approach, and places her work in the auction tradition, but her findings are consistent with
supply function models. Her main results are that the larger generator in the British duopoly
had higher bids, relative to its marginal costs, and that all stations tended to submit higher
bids when their owner had more lower-cost capacity available. Raising the price bid by one
station will raise the price received by all the stations that are infra-marginal when it runs.
The more of these that belong to the station’s owner, the greater the incentive to bid up its
price.
Most papers on other markets have followed the approach of Cournot competition, in
which generators decide how much of their capacity to make available to the market in each
period, and prices are set to bring demand into balance with this supply. Cournot models
have been used to analyse markets in Sweden (Andersson and Bergman, 1995), the Nordic
countries (Amundsen et al, 1999) and California (Borenstein and Bushnell, 1999), amongst
others. These papers attempt to predict energy prices, and might be construed as in the
“price-increase” tradition, but the strategic instrument is in fact the capacity offered to the
5
market. Since each generator assumes that its rivals will not respond to any reduction in its
own capacity, the equilibria tend to involve greater reductions in output, and higher prices,
than supply function models in which generators expect their rivals’ output to increase when
they try to exercise market power.
Three papers clearly look at the generators’ decision on whether to offer their
capacity to the market. Newbery (1995) showed that the capacity payment system gives
British generators an additional incentive to keep some of their capacity from the market, and
quantified the size of this incentive. Anecdotal evidence suggests that generators have
sometimes delayed the end of an outage if this would raise the level of capacity payments.
Crampes and Creti (2001) show how generators would have an incentive to reduce the
amount of capacity that they offered into the market if this would then turn the market from
one with excess capacity and low bids into a capacity-constrained market with high prices in
equilibrium. Wolak and Patrick (1997) introduced the idea of withholding infra-marginal
capacity in order to raise prices without raising their bids above marginal cost. They show
that the generators declared roughly the same amount of capacity declared available, relative
to demand, in peak (November – February) and off-peak months (the remainder) between
April 1991 and March 1995. Since demand was much lower in off-peak months, this meant
that a large amount of capacity was not made available to the market at those times. The
mean availability in peak months was 51 GW, while the mean availability in off-peak months
was 43.8 GW. Furthermore, National Power and PowerGen had lower availability figures,
for each type of plant that they owned, than comparable plants in North America, when new
entrants to the market were achieving higher availabilities than comparable American
stations.
Wolak and Patrick suggest that the generators would prefer to withhold capacity from
the market, compared to raising their bid prices, because capacity withholding could be
disguised as maintenance, while the regulator could easily compare each station’s bid price to
an estimate of its marginal costs. In practice, however, the regulator has not often referred to
explicit comparisons of a station’s bid with its costs. A study on constrained-on plant (Offer,
1992) gave revenue and cost figures for a number of stations that had submitted high bids to
exploit their geographical position, and a report on gas turbine plant also considered station
costs. There have been a number of comparisons of differing bids made by the same station
at different times, with the implication that at least some of them could not have been costreflective.
However, the overall level of prices has rarely been related to the generators’
costs – a study on “costs and margins” promised in 1993 never saw the light of day, and the
regulator has not referred to it since.
In contrast, however, the major generators were required to provide the regulator with
an annual forecast of their capacity availability, on a set-by-set basis, a half-way update, and
a reconciliation at the end of each year. The licence condition governing these statements
made no attempt to lay down levels of availability, but they provided the regulator with all
the evidence that he might need, were he to investigate a generator for anti-competitive
behaviour. The regulator has used information on forced outage rates and demand adjusted
6
to normal weather conditions to determine whether a temporary reduction in the plant
margin, which caused high capacity payments in January 1995, was “abnormal” (Offer,
1995). The major generators also had to provide the regulator with advance notice of any
plans to close capacity, and consider the option of offering it to sale to a third party. The
regulator could appoint an independent assessor to decide whether the plans are reasonable,
and it is unlikely that an assessor would approve closures when capacity payments are
predicted to be high.6
All in all, the generators may have had less freedom with their
capacity declarations than with their price bids.
4. Capacity withdrawal and capacity payments
In some circumstances, keeping capacity away from the market is a perfectly acceptable
competitive response. A peaking station is unlikely to be required to run overnight, when
demand is low. The station may be technically available, in that it is not undergoing
maintenance, but it would not make commercial sense to staff the station at operational levels
– the staff would be wasting their time, waiting for a call from the grid operators that is not
going to arrive. The station should only be made available in the sense of being staffed and
ready to generate during daytime hours when it is likely to be needed. Availability figures in
the Pool were based on this definition, and some mid-merit and peaking stations regularly
declared themselves “commercially unavailable” at off-peak times.7
If this is done without
affecting prices, then it can be counted as a proper competitive response. This would be the
case when there the chance that the station would actually be required to run is negligible.8
If
prices rise in response, however, then there is a possibility of market manipulation. The
question is whether the price before the station was withdrawn was high enough to cover the
costs of making it available over the relevant period. If the price was above the avoidable
cost of keeping the station open, the “competitive” option would be to make it available,
while if the price was below this level, a single station’s owner would lose money by keeping
the station in the market, and should withdraw it.
This gives us a test. If generators had perfect foresight and acted competitively,
capacity payments should equal the avoidable cost of a peaking station. Higher capacity
payments would imply anti-competitive behaviour. It is normal to compare costs and
revenues over the course of a year, and generators do recover most of their costs at the time
6
It is not strictly necessary to use predictions, since companies trade annual contracts to hedge capacity
payments, and so actual market prices can be used to compare a station’s costs with its potential revenues.
7
Officially, stations in the US would still be considered available in these circumstances (Personal
Communication: M. Curley of NERC, 29 August 2000), and so comparisons between US and UK availability
data can be misleading – US stations should only be recorded as unavailable when shut down for, or awaiting,
maintenance. In practice, Joskow and Schmalensee (1987) show that units with a low ratio of actual to
available output tended to have lower availability figures. They ascribe this to the strain of frequently turning
the station on and off, leading to increased maintenance, but the pattern is consistent with stations being
declared unavailable simply because they do not expect to run.
8
The calculation behind capacity payments actually uses every station’s highest availability over the previous
eight days, and so declaring a station unavailable overnight and at weekends should have no impact on the
capacity payment.
7
of the system peak. That does not imply that capacity payments should be zero at all other
times, however. Even in off-peak months, some plant must be made available, but not
expected to generate, at the daily peak, and will incur an avoidable cost in doing this.
Capacity payments must be expected to exceed zero, if only slightly, to make it worthwhile to
declare this reserve capacity available.
At the annual level, the Monopolies and Mergers Commission (1996) has reported
that the fixed costs of an open cycle gas turbine plant were £6/kW per year. This might be a
lower bound on the expected capacity payment that a competitive generator would require in
order to keep its station open. An independent assessor’s report on plant closures (Offer,
1998) gave a fixed cost for a unit of £22.50/kW per year, with an extra £5/kW per year for
station costs. However, the unit cost (for a 200 MW unit) included £8 million for an
overhaul, spread over 4 years of operation, or £10/kW/year. For a station close to the end of
its life, but capable of continued running without such an overhaul, the avoidable cost would
be £17.50/kW per year. In 1992, Offer’s report on constrained-on plant included output and
cost figures for a number of stations – with appropriate assumptions on fuel prices, thermal
efficiencies, and variable operations and maintenance costs, the implied fixed costs were
around £25/kW per year. That might form an upper bound for the expected capacity payment
that should trigger plant closures.
These bounds should be compared with the expected capacity payments at the time
closure decisions would be made, which are unobservable.9
The actual capacity payments
will be higher than the expected level if the weather is cold (England and Wales have a
winter peak) or plant failures are above average, and lower if a warm winter coincides with
few plant failures. Over a number of years, however, the average actual capacity payment
might form a good proxy for the expected level. Table 1 shows the capacity payment in each
year from 1990/1 onwards. Figures are given for a complete financial year, and broken down
into monthly averages for the winter, the summer, and the “shoulder” months (March, April,
September and October). Competitive behaviour would imply that the annual average prices
were close to the range of £6/kW to £25/kW discussed above, and that these should be
concentrated in the winter months. Capacity payments in the summer and the shoulder
months should be low (but not zero) if the generators are acting competitively.
The annual capacity payments are volatile, and there are a number of individual years,
particularly later in the period, when they exceed our range for the competitive level. Their
cumulative average over the life of the Pool, however, does remain below the upper bound of
competitive values. In most years, capacity payments outside the winter months are also low
enough to be compatible with a competitive outcome. In the following paragraphs, we
comment on some of the payments that appear to have exceeded competitive levels.
9
Some contracts for differences were written to hedge capacity payments alone, and would form a guide to
participants’ expectations, but prices for these are not in the public domain.
8
Table 1: Capacity payments in England and Wales, 1999-2000 prices
Monthly capacity payment, £/kWAnnual Capacity
Payment, £/kW
Average,
1990/1 to date,
£/kW
Winter
(Nov-Feb)
Shoulder Summer
(May-
Aug)
1990/1 0.55 0.55 0.10 0.00 0.02
1991/2 13.73 7.14 1.59 1.24 0.26
1992/3 1.76 5.35 0.23 0.16 0.01
1993/4 2.89 4.74 0.51 0.09 0.05
1994/5 31.49 10.09 6.53 0.45 0.12
1995/6 42.59 15.50 8.21 0.76 0.64
1996/7 30.32 17.62 3.45 2.19 1.75
1997/8 7.59 16.37 1.45 0.20 0.07
1998/9 8.82 15.53 0.94 0.26 0.96
1999/00 24.10 16.39 2.18 1.90 1.94
2000/01 38.08 18.36 1.11 5.94 2.47
The table suggests that capacity withholding to increase capacity payments was not a
significant feature of the early years of the Pool. The one, quite well known, exception
concerns behaviour by PowerGen in the summer and early autumn of 1991. Each day, the
company declared that some of its stations would not be available on the following day,
raising the loss of load probability, and setting capacity payments at a higher level than if the
stations had been available. The stations were then re-declared to be available, and collected
the capacity payment, which had been fixed and could not be re-calculated. The company
soon stopped the practice, in response to criticism by the regulator, who imposed the
availability reporting requirements described above. Furthermore, the Pool Rules were
changed to use every station’s highest availability (declared or actual) over an eight-day
period when calculating the loss of load probability. This made PowerGen’s original tactics
effectively impossible – a station would have had to be unavailable for a week before
delaying its return to service would affect capacity payments. If a plant has been on outage
for eight days or more, however, a company could gain from delaying its return, and there is
plenty of anecdotal evidence that this sometimes happened. In the first few years of the Pool,
however, any extended outages of this kind failed to raise capacity payments above
“competitive” levels.
In the period since 1994/5, however, capacity payments were much higher, averaging
more than £20/kW. The high payments in 1994/5 were the subject of an Offer investigation,
which found them to be the result of an unlikely combination of plant outages, as discussed
above. In 1995/6, payments were concentrated in the winter, when demand was high, and
raised by exports to France (where output was reduced by strikes). In other words, they
could once again be consistent with competitive availability choices.
9
The following year, however, capacity payments were at similar levels (on average)
throughout the year. That might suggest that generators were making enough plant available
at the winter peak, but withdrawing too much of it at other times of the year, raising “off
peak” capacity payments. The data for 1999/00 would be consistent with this, but Ofgem
(1999) provides an alternative explanation. Capacity payments depended not only on the
amount of plant on the system, relative to demand, but also on its reliability. That was
measured in a way which tended to make new capacity appear less reliable, and so capacity
payments would be higher, the greater the proportion of new plant on the system. The plant
margin was actually slightly higher in July 1999 than in July 1998, but capacity payments
were five times as high. Part of this was due to an increase in the proportion of new plant:
half of the increase was due to the way in which a single (lengthy) outage at a relatively new
nuclear station affected its calculated reliability (Ofgem, 1999, pp 25-6). This begs the
question of why the payments were calculated in this way, but also suggests that the high
payments in 1999/2000 were “innocent” rather than the result of strategic capacity
withdrawal.
In the summer of 2000, however, Ofgem took action when Edison Mission Energy
withdrew 480 MW of capacity from the market. Edison was one of the companies that had
accepted Ofgem’s short-lived market abuse licence condition, and the regulator started an
investigation into the company’s decision. Ofgem (2000) reported that “on the basis of
analysing the avoidable costs of the capacity withdrawn by the company … against spot and
forward prices it appears that Edison may be in breach of the market abuse condition.”
Edison then agreed to return the plant to service, and Ofgem took no further action. This is
the first time that the regulator has taken action against a generator for withdrawing capacity
since PowerGen’s blatant manipulation of 1991. The highest prices of 2000/1 came two
months later, in September 2000, and Ofgem’s investigation once again concluded that the
shortcomings of the capacity payment’s calculation, rather than deliberate action on the part
of generators, was to blame.
While the capacity payments mechanism had serious flaws in its calculations, and
generators did engage in “tactical” withdrawals of capacity, it is probably reasonable to
conclude that for most of the 1990s, generators did not “strategically” withhold enough
capacity from the market to force expected capacity payments above competitive levels.
While the level and pattern of capacity payments is generally consistent with
“competitive” behaviour, the way in which the dominant generators withdrew capacity from
the market does deserve a comment. In general, it is likely to be more expensive to retain
two stations, closing part of the capacity at each, than to close a single station.10
Furthermore, if a station is to be closed, making the closure permanent at once will allow the
company to dispose of the site. National Power and PowerGen, however, have frequently
closed stations in stages, and mothballed plant before announcing its permanent closure.
Mothballing plant might have been part of an entry-deterring strategy, with the implied threat
10
Keeping part of a station open can be justified if it is required to provide system support, perhaps by easing a
transmission constraint, and can earn enough from doing so to offset the additional costs involved.
10
that the market could be flooded with spare capacity, should there be excessive entry. In
practice, however, such threats were never carried out, and appear to have been very
unsuccessful in deterring entry. Closing a station in stages will sometimes be a tactical
response to the regulator’s requirement that the generators should consider selling stations
that they plan to close. It is practically impossible to buy part of a station, and so partial
closure allows the generators to avoid selling plant that they believe is surplus to
requirements. If the closure was justified on “competitive” grounds, then a stand-alone
operator should not be able to make a profit from the station, but the generators may have
feared that a rival would attempt to force the incumbents to withdraw other plants,
accelerating their loss of market share. If the closure was not justified on competitive
grounds, but was part of an anti-competitive capacity withdrawal, then the plants would be
more attractive to rival generators, who could make a profit without further closures from the
incumbents. The evidence above, however, and the regulator’s investigations into proposed
closures, suggest that the level of closures was generally justified on competitive grounds,
even if their manner was in some respects dubious.
5. Capacity withdrawal to raise energy prices
The previous section has considered withdrawing high-cost capacity in order to boost
capacity payments. A second kind of manipulation has been suggested by Wolak and Patrick
(1997). An incumbent might withhold low-cost capacity, ensuring that higher-cost stations
would set the system marginal price, thus raising it, without requiring any station to bid
significantly above its costs. This strategy might be undertaken at times of low demand, and
the stations made available when demand was higher, so that the system marginal price could
be increased without increasing capacity payments. Wolak and Patrick suggest that it would
be hard for a regulator to detect this kind of capacity withholding, since it could be disguised
as maintenance.
Withholding a station in this way, however, will reduce its load factor. The stations
that run instead will have higher load factors. This section of the paper uses information on
load factors before and after restructuring to construct a test for this kind of capacity
withholding. The CEGB regularly published the load factors of its stations, while the MMC
published information on the load factors of National Power and PowerGen’s coal-fired
stations in its 1996 reports into their merger proposals. NGC provided load-duration curves,
showing the number of hours in each year for which demand above a given level.
If all stations were available throughout the year, and there were no transmission
constraints or other operating restrictions, then it would be possible to match the generators’
output exactly to the shape of the load-duration curve. The most efficient stations would
have a load factor of 100%, while less efficient stations would run for as long as there was
demand for their output. In practice, however, it is extremely unlikely that a station would be
available for the whole year, and so its load factor will be less than 100%. The “missing”
output must be made up by another station, and will be outside the load-duration curve.
11
Operating restrictions will further reduce the output of the more efficient stations, and
increase the output of those with higher costs. This will give us the pattern shown in figure 1.
The line is the load-duration curve, showing the demand for each hour in the year,
ranked in decreasing order of demand. The highest demand is placed at hour 0, and the
lowest at hour 8760 (8784 in a leap year, and 8640 in 2000/1, when the Pool was abolished
on March 26). The rectangles show each station’s output, ranked in merit order – the stations
with the lowest operating costs, which ought to run the most, are at the bottom. The dark
grey portions show the output that was “within” the load-duration curve, and would have
been produced if generation had perfectly matched demand. The light grey portions are the
output that was only required because some other stations were generating less than the
system could have accepted from them – the dark grey areas do not fill the area under the
load-duration curve. The total area of the light grey portions should be the same as the area
of the blank spaces under the load-duration curve.
The CEGB’s last published output data were for 1985/6, 1986/7 and 1987/8. Output
patterns in 1984/5 were significantly disrupted by the miners’ strike of that year, and earlier
years might be increasingly unrepresentative of the system as it was restructured. We assume
that the CEGB aimed to run its stations in the “true” merit order, and derive this from the
mean load factor of each station across the three years – the coal-fired stations with the
highest load factors during this period are assumed to be highest in the merit order.
Throughout the 1990s, the price of coal in the UK was sufficiently high, relative to
the price of gas, that CCGT stations were always above coal-fired stations in the merit order,
given their greater thermal efficiency. To obtain the merit order for the 1990s, we simply
inserted the CCGT stations above the best coal-fired stations, with the newest stations at the
top. This assumption would break down for 2000/1, however. In the second half of that
year, the price of gas delivered to power stations was 50% higher, relative to the delivered
price of coal, than over the previous eighteen months. This was the only time in the Pool’s
history that a CCGT would have had significantly higher fuel costs than a coal-fired station.
Table 2 shows how this affected the type of plant which was setting SMP – in 1999/2000,
CCGTs were only on the margin for one percent of the time. In the first half of the following
year, this had risen to 4.8 percent of the time, and 13.5 percent of the time in the second half
of the year. Given the change in fuel prices, this was clearly due to a change in the costbased
merit order, and not to a change in generators’ strategies. Since the merit order
changed in the middle of the year, it is not possible to compare 2000/1 with the years of a
stable merit order. Results for each year of the Pool are reported, but the comparisons are
based on the ten years from 1990/1 onwards.
12
Table 2: Plants setting System Marginal Price, by fuel type (number of half-hours)
1999/2000 April – Sept 2000
Oct 2000 – Mar
2001
Coal 14,260 81.2% 6,861 78.1% 5,416 63.7%
Interconnector 2,336 13.3% 843 9.6% 1,592 18.7%
PSB 757 4.3% 652 7.4% 339 4.0%
CCGT 170 1.0% 422 4.8% 1,147 13.5%
OCGT 43 0.2% 0 0.0% 0 0.0%
CCGT/CHP 2 0.0% 5 0.1% 0 0.0%
No SMP was set 0 0.0% 1 0.0% 2 0.0%
Totals 17,568 100% 8,784 100.0% 8,496 100.0%
Returning to figure 1, we could simply take the ratio of the blank area under the loadduration
curve to total generation as an indicator of the “mis-match” between generation and
demand, and compare this across years. This would ignore the role of forced outages,
however. The CEGB (along with most other system planners) assumed that a proportion of
plant would be unavailable at any time, for technical reasons. This means that we should
never expect to fill the area under the load-duration curve with generation. We can get a
more realistic “ideal” pattern if we shrink the load-duration curve along the horizontal axis,
while stretching it along the vertical axis to preserve the total area. Using a 10% forced
outage rate (the CEGB norm) the maximum length of running expected would shrink to 90%
of the year (7884 hours), while a peak demand of 45 GW would be raised to 50 GW. Figure
2 shows the dotted original load-duration curve, and the revised curve. A greater proportion
of the generator’s output is under this revised curve, reflecting the fact that the industry needs
enough capacity to meet the peak demand plus a margin for unavailable stations.
We will make a second adjustment, to take account of new stations, and the nuclear
stations. Once they have been commissioned, most new stations (and all built during our
period) would be at the top of the merit order, running nearly continuously. While they are
commissioning, however, new stations will only produce intermittently, to a timetable
dictated by their testing requirements, rather than by market conditions. The amount of plant
being commissioned varied over our period, and it would be unreasonable to count output
that was “lost” during commissioning as evidence of capacity withholding. Similarly, while
nuclear stations attempted to run at high load factors, they all too often failed to achieve
these, for reasons that had nothing to do with National Power and PowerGen (which did not
own any nuclear stations). The striped area in figure 2 represents the output lost due to plant
commissioning or poor nuclear performance, compared to a target load factor of 85%. Note
that this still allows for some “under-performance” compared to the vertical section of the
adjusted load-duration curve, which is at a load factor of 90%.
13
Table 3 shows how these calculations work in practice, for four stations in 1990/91.
Hinkley Point B was a nuclear station which achieved a load factor of 76.5%, and output of
7,510 GWh. At the top of the merit order, we assume that a “perfect” performance would
have been a load factor of 90%, and an output of 8,830 GWh. The difference between 8,830
GWh and 7,510 GWh is a measure of the lost output due to its failure to match the loadduration
curve – some 1,320 GWh. Since Hinkley Point B was a nuclear station, however,
we discount shortfalls below a load factor of 85%, and so only 490 GWh (8,830 – 8,340) are
treated as truly lost.
Table 3 – illustrative calculations for selected stations, 1990/91
Station
Hinkley
Point B Drax Cottam Drakelow C
Capacity 1120 3750 1920 448
Actual Load Factor 76.5 79.1 76.1 25.6
LDC load factor 90 90 84.3 15.1
Output at this LF 8,830 29,565 14,179 593
Output at 85% LF 8,340
Actual Output 7,510 25,994 12,805 1,006
Lost Output 1,320 3,571 1,374 0
Truly Lost Output 490 3,571 1,374 0
The large coal-fired station at Drax was also high enough up the merit order to have a
potential load factor of 90%. Its actual load factor was 79.1%, and the shortfall was 3,571
GWh (29,565 – 25,994), all of which was counted as “truly lost”. Another large coal-fired
station, Cottam was slightly lower down the merit order, and had a potential load factor, had
all stations above it been operating at their maximum load factors, of 84.3%. Instead, its load
factor was 76.1%, and so it accounted for 1,374 GWh of lost (and “truly lost”) output.
Finally, Drakelow C was a low merit station with small, less efficient, units. Its potential
load factor would only have been 15.1%, had all the stations above it been operating at their
maximum load factors, but their shortfalls actually allowed it a load factor of 25.6%. With an
actual output greater than the amount under the load-duration curve, it was not responsible
for any lost output.
In general, the adjustment for new plant is made for the first financial year in which a
new station generated. Some stations started their commissioning late in the year, however,
and did not have time to produce much output before the year end. It would be inappropriate
to discount a large amount of lost output on account of a plant that was only just starting the
commissioning process, however. Therefore, new stations with load factors of less than 25%
were moved down the merit order to be treated as peaking plant, where they would not affect
the amount of lost output. The adjustment for output lost in commissioning was made in the
14
following year. Taking the two years together, this will reduce the amount of lost output that
is discounted due to the plant’s commissioning.
Many new plants had several units, but they were generally commissioned
simultaneously, or nearly so, and therefore treated together. The main exception is the Drax
coal-fired station, which commissioned its last two units (out of six) in 1985 and 1986.
Drax’s reported output is not broken down by unit, and any output produced by the new units
before they were formally commissioned was not assigned to the station, but reported as a
total for all such units, (which included several nuclear stations at this time). Since 2.5 GW
(3.1 GW) of capacity had already been commissioned before 1985/6 (1986/7), it would be
inappropriate to apply the adjustment for new plant to the entire station, when only one 625
MW unit was being commissioned in each year. We therefore assumed that the
commissioning unit achieved a load factor of only 25%, and deduct 3.3 TWh (625 MW ×
60% × 8,760 hours) of output from the calculated amount of lost output in 1985/6 and
1986/7. This almost certainly over-estimates the amount of output to be discounted on behalf
of these units, improving the performance of the CEGB. Since the post-privatisation industry
turns out to be better than the CEGB at fitting its output to the load-duration curve, any
assumption that improves the CEGB’s performance will narrow the gap between the two.
This adjustment also moves Drax to the top of the merit order for coal-fired stations, based on
the implied performance of its “established” units.11
Two post-privatisation stations, South
Humber Bank and Seabank, were also commissioned in two stages, and each part is treated as
a separate station in this analysis.
This analysis assumes that the shape of the load-duration curve is the only constraint
on generators. In practice, some stations have to reduce their output because of constraints
on the transmission system. Output from low-cost stations has to be replaced by output from
higher-cost stations that are closer to the load centres, in order to avoid overloading the
transmission lines from the more distant stations. Transmission constraints rose significantly
in the early 1990s, in part because of the amount of new plant being connected to the
transmission system and old plant being disconnected. If this meant that high-merit stations
had to generate less, it might be expected to worsen the fit between the load-duration curve
and actual generation. This would provide an “innocent” explanation for any deterioration in
performance in the early 1990s. In practice, we will see that this fit improved after
privatisation, despite any impact of constraints.
Our test, therefore, is to measure the blank area under the adjusted load duration
curve, as a proportion of total output. The shaded area above the load-duration curve can be
denoted “excess output”, the striped area under the load-duration curve is the part of this that
is due to low output levels from nuclear plant or new entrants, and the blank area under the
load-duration curve is “truly excess output”. If generators are withholding capacity that
11
Drax was also the first station to have Flue Gas Desulphurisation equipment fitted, and while this slightly
reduced its thermal efficiency, the Environment Agency required National Power to run Drax in preference to
other stations without FGD, reinforcing its position at the top of the coal-fired merit order.
15
should have been in merit, in order to raise the system marginal price, the amount of truly
excess output should rise after restructuring. Table 4 gives the figures.
Table 4: Output and the load-duration curve
Year Total Output
(TWh)
Within load-
duration
curve
Lost new and
nuclear
output
Truly excess
output
(TWh)
Truly excess
%
1985/6 231,983 200,269 12,801 18,913 0.082
1986/7 237,785 192,715 21,359 23,711 0.100
1987/8 243,912 206,643 13,940 23,330 0.096
1990/1 261,537 224,515 16,730 20,291 0.078
1991/2 263,259 232,226 13,392 17,641 0.067
1992/3 262,655 229,947 14,538 18,171 0.069
1993/4 266,523 229,584 14,049 22,890 0.086
1994/5 268,953 227,025 21,192 20,736 0.077
1995/6 283,779 245,138 23,756 14,885 0.052
1996/7 286,851 249,705 12,214 24,931 0.087
1997/8 288,193 252,083 10,870 25,239 0.088
1998/9 293,368 254,962 10,773 27,633 0.094
1999/00 296,741 252,269 23,723 20,749 0.070
2000/1 306,083 245,925 23,596 36,562 0.119
On this definition, 9.2% of output was “truly excess” in the three years of CEGB operation.
In almost every year after the restructuring, the proportion of truly excess output was lower.
The one glaring exception is 2000/1, the last year of the Pool, and the way in which the
changing fuel prices during the year make the comparison inappropriate is discussed above.
We can test the hypothesis that the mean percentage of truly excess output was the
same in the first three years and the next ten years of data, using Excel’s built-in t-test. If we
assume that the two samples have the same variance, the p-value of a two-tailed test for equal
means is 0.075, while if we assume unequal variances, it is 0.079.12
In either case, we reject
the hypothesis of equality at the ten percent level. After privatisation, the generators were
significantly better at fitting their output to the load-duration curve. If they had been
withholding output in order to raise prices, this would have worsened their fit to the loadduration
curve.
12
The variances are 9.11E-05 and 1.55 E-04, which are hardly equal, but an F-test does not reject the
hypothesis of equality at the 10% level.
16
Does this result depend upon the values chosen for forced outages and the target load
factor for new and nuclear plant? Table 5 shows how the p-value for the hypothesis of equal
means varies with these parameters. The table always reports the greater of the two values
for equal and unequal variances.
Table 5: p-values for testing the hypothesis that the proportion of truly excess output did not
change between 1985/6-7/8 and 1990/1-99/00
Availability after forced outagesTarget load factor for
new and nuclear plant 1.00 0.95 0.90 0.85 0.80
1.00 0.015
0.95 0.023 0.016
0.90 0.037 0.028 0.044
0.85 0.056 0.046 0.079 0.174
0.80 0.067 0.055 0.094 0.153 0.632
0.75 0.059 0.048 0.074 0.097 0.334
0.70 0.049 0.041 0.058 0.074 0.147
0.65 0.040 0.034 0.046 0.058 0.089
If the rate of forced outages is 0.1 or lower, then we continue to get a significant difference in
the proportion of truly excess output, at the ten percent level. If the rate of forced outages is
greater than this, and the target load factor for new and nuclear plant is close to that expected
for established conventional plant, the difference is not significant at this level. None of the
tested parameter values implied that the CEGB had a better performance than the privatised
industry, however. Furthermore, the qualitative results do not depend upon the treatment of
new plant. If we simply look at the proportion of output under the load duration curve, with
no adjustment for nuclear outages or new plant (apart from continuing to ignore those with
load factors under 25% in their first year), it rises from 0.840 under the CEGB to 0.864 in the
first ten years of the Pool. The improvement is significant at the five percent level if we
assume equal variances, but not significant (p = 0.252) if we assume that the variances are
unequal.13
6. Conclusion
We have studied two ways in which generators could play “Cournot” strategies, raising
prices by withholding capacity from the market. From the evidence available, neither
strategy seems to have been significant in England and Wales. After an initial abuse of the
capacity payment, the regulator has required the major generators to provide regular
information on their capacity and its availability. It appears that the companies have
13
The variances are 7.33 E-04 and 1.40 E-04 respectively. An F-test for their equality is not rejected at the 5%
level, but is rejected at the 10% level.
17
generally made capacity available in a competitive manner, declaring plant whenever it is
technically available and can expect to earn more than its avoidable costs. Anecdotal
evidence suggests that an outage will sometimes be prolonged unnecessarily with the
intention of keeping capacity payments high, and there were times near the end of the Pool’s
life when the regulator found that individual companies had acted in an anti-competitive
manner. For much of the period, however, the overall level of payments implies that such
practices were not used to raise prices by a significant amount. The industry’s overall
performance at fitting its output to the load-duration curve improved significantly after
privatisation, when systematic capacity withholding would have worsened it.
Cournot models of the electricity industry do have attractive features. They can
support detailed cost modelling, and do not usually suffer from the multiple equilibria
common with other modelling approaches. Against this, however, they do not give a good
representation of the way in which prices are set in electricity markets, and their price
predictions have generally been too high.14
This paper suggests that models in which firms
set prices, rather than quantities, are likely to give better results.
14
A rare exception is Bushnell (2003), which shows that a Cournot model, updating the assumptions of
Borenstein and Bushnell (1999), gives a good fit to the prices experienced in California in the summer of 2000.
While this is before the final crisis, the market was hardly performing well at this stage.
18
References
Amundsen, E., L.Bergman and B.Andersson (1999) Trade and Market Power on the
Northern European Market for Electricity, mimeo, Stockholm School of Economics
Andersson, B., and L.Bergman (1995) “Market Structure and the Price of Electricity: An Ex
Ante Analysis of the Deregulated Swedish Electricity Market” The Energy Journal
vol 16 no 2, pp 97-109
Borenstein, S. and J.Bushnell (1999) “An Empirical Analysis of the Potential for Market
Power in California’s Electricity Market.” Journal of Industrial Economics, Vol 47, No.
3, pp 285-323
Bushnell, J. (2003) Looking for Trouble: Competition Policy in the U.S. Electricity Industry
CSEM WP-109, University of California Energy Institute
Crampes, C. and A.Creti (2001) Price Bids and Capacity Choice in Electricity Markets, mimeo,
Institute d’Economie Industrielle, University of Toulouse
von der Fehr, N-H M., and D. Harbord (1993) “Spot Market Competition in the UK Electricity
Industry”, Economic Journal, vol 103 no 418 pp 531-46
Green, R.J. (1994) “Britain’s Unregulated Electricity Pool” in (ed) M.Einhorn From Regulation
to Competition: New Frontiers in Electricity Markets, Boston, Kluwer
Green, R.J, and D.M.Newbery (1992) “Competition in the British Electricity Spot Market”
Journal of Political Economy vol 100 no 5, pp 929-53
Hogan, W.W. (2000) Making Markets in Power Cantor Lecture to the Royal Society for the
Encouragement of Arts, Manufactures and Commerce, London
Joskow, P.J. and R.Schmalensee (1987) “The Performance of Coal-Burning Electric
Generating Units in the United States: 1960-1980”, Journal of Applied Econometrics,
vol. 2, no. 2, pp 85-109
Kreps, D. and J.Scheinkman (1983) “Quantity Precommitment and Bertrand Competition Yield
Cournot Outcomes”, Bell Journal of Economics, vol 14 pp 326-337
Monopolies and Mergers Commission (1996) National Power PLC and Southern Electric plc:
A report on the proposed merger, Cm 3230, London, HMSO
National Audit Office (2003) The New Electricity Trading Arrangements in England and
Wales: Report by the Comptroller and Auditor General, HC 624 of Session 2002-2003,
London, The Stationery Office
Newbery, D.M. (1995) “Power Markets and Market Power”, The Energy Journal, Vol 16, No
3, pp 39-66
Newbery, D.M. (1998) “The Regulator’s Review of the English Electricity Pool” Utilities
Policy, vol 7 no 3, pp.129-142
Offer (1991) Pool Price Enquiry Birmingham, Office of Electricity Regulation
Offer (1992) Report on Constrained-On Plant, Birmingham, Office of Electricity Regulation
Offer (1995) Generators’ Pool Price Undertaking 1994/95 Birmingham, Office of Electricity
Regulation
Offer (1998), Independent Assessor’s report on plant closure. 17 April 1998 Birmingham,
Office of Electricity Regulation
19
Ofgem (1999) Rises in Pool prices in July: a decision document. October 1999 London,
Office of Gas and Electricity Markets
Ofgem (2000) Ofgem’s investigation of Edison First Power under the market abuse licence
condition: Initial findings, London, Office of Gas and Electricity Markets
Wolak, F. and R.H.Patrick (1997) The Impact of Market Rules and Market Structure on the
Price Determination Process in the England and Wales Electricity Market Power
Working Paper PWP-047, University of California Energy Institute
Wolfram, C.D. (1998) “Strategic bidding in a multiunit auction: an empirical analysis of bids to
supply electricity in England and Wales”, Rand Journal of Economics, vol 29, no 4, pp
703-725
Wolfram, C.D. (1999) “Electricity Markets: Should the Rest of the World Adopt the United
Kingdom’s Reforms?” Regulation, Vol. 22 (4): pp.48-53
20
Figure 1: The Raw Load-Duration Curve
MW
8760 Hours
MW
8760 Hours
Figure 2: The Adjusted Load-Duration Curve
21